Research Article
Object Detection Based on Swin Deformable Transformer-BiPAFPN-YOLOX
Table 1
Results of ablation experiments using different components on the COCO 2017 validation set. Swin DeTr denotes Swin Deformable Transformer.
| Methods | AP | AP50 | AP75 | APS | APM | APL | Param (M) | GFLOPs | Infer time (ms) | FPS |
| DarkNet53-PAFPN-YOLOX [2] | 47.4 | 67.3 | 52.1 | 27.5 | 51.5 | 60.9 | 63.7 | 185.3 | 11.1 | 90.1 | Swin transformer-PAFPN-YOLOX | 48.4 | 67.8 | 52.6 | 29.3 | 52.6 | 61.8 | 86.9 | 221.6 | 14.5 | 69.0 | Swin DeTr-PAFPN [26]-YOLOX | 49.1 | 68.3 | 53.0 | 30.7 | 54.6 | 62.9 | 53.2 | 173.1 | 12.4 | 80.6 | Swin DeTr-BiPAFPN-YOLOX | 49.7 | 68.6 | 53.2 | 31.2 | 54.9 | 63.3 | 44.5 | 110.7 | 10.1 | 98.7 |
|
|